Prognostic indicators of survival in ALS

Published reports of median length of survival in ALS patients vary from 2 to 4 years.^1-5 Factors identified as associated with survival include patient age and site of disease onset, especially bulbar onset of ALS.^3-5 Haverkamp et al.,¹ in a study with extended follow-up of a large cohort of ALS patients, identified the rates of change in the Appel ALS score and its respiratory subscore as prognostic factors in a proportional hazards model of survival in ALS. Their model also included the time from onset of symptoms to diagnosis of ALS. However, baseline Appel score values did not predict survival. We also found an association between survival in a 9-month clinical trial and rate of deterioration in functional abilities assessed using the ALS Functional Rating Scale (ALSFRS) but not baseline ALSFRS value.⁶

Several reports showed that indices of pulmonary function are most strongly predictive of survival in ALS. Bensimon et al.⁷ identified percent-predicted forced vital capacity (FVC%) as a prognostic factor for survival in a randomized clinical trial of riluzole in ALS. Ringel et al.² found rate of decline in pulmonary function(defined as the slope of a pulmonary megascore, 50% of which is attributable to FVC%) to be associated with length of survival, whereas Fallat et al.⁸ reported that patients who survived to a common censoring date had higher FVC values at the beginning of the observation period than those who had died. Fallat et al.⁸ commented on the nonlinear pattern of decline in pulmonary function and remarked that gas exchange is well maintained until FVC% declines to less than 20% of predicted. Several authors have commented on the lack of identifiable association between last measured FVC% values and death, even in as uniformly fatal a condition as ALS.

We sought to evaluate prognostic variables for death in patients with ALS to determine which factors can be used in selecting patient cohorts for the conduct of clinical trials and to find practical indicators that clinicians might use in guiding their day-to-day management of ALS patients. In addition, because of the nonlinearity in individual patient's pulmonary function declines and the variability noted in terminal FVC% measures, we evaluated patients' acid-base balance as a secondary indicator of the systemic metabolic consequences of declining pulmonary function. We now report that metabolic indicators of the degree of compensation for existent pulmonary insufficiency in the months before death may supplement the value of pulmonary function measures as predictors of outcome in ALS clinical trials.

Methods. We report observations based on the placebo group of 245 patients enrolled in the ALS CNTF Treatment Study Group (A.C.T.S.) study, which sought to determine the efficacy of recombinant human ciliary neurotrophic factor (rhCNTF) for the treatment of ALS. The results of the full trial have already been reported.⁹ The present report concerns only th5e outcome among placebo patients because patients treated with rhCNTF developed side effects, including some affecting respiratory and bulbar function.⁹

Patients interested in participating in the A.C.T.S. study were first entered into a 2- to 6-month screening protocol. The treatment trial itself enrolled 730 patients with probable or definite ALS, a score on the ALSFRS of≥18, no complicating medical diseases, who did not require mechanical ventilatory support for any part of the day regardless of FVC% values. The patients subsequently underwent monthly evaluations of physical status and ALS progression.⁹ Measurements included muscle strength, pulmonary function, and activities of daily living testing; vital sign and clinical laboratory monitoring; and adverse event reporting. Patients who completed the 9-month treatment trial were permitted to enroll in an open-label follow-up study of rhCNTF. Vital status was assessed both at the end of the blinded trial and at the end of the open-label study. Only data from the lead-in and placebo treatment phases of the study were incorporated into the evaluation of survival prognostic factors.

Statistical analyses were performed using JMP Version 3.1 (SAS Institute Inc., Cary, NC). Survival data were analyzed using the Cox proportional hazards model, a multivariate regression method that examines the effect of presumed prognostic factors on survival time. Survival beyond a specified time is expressed in terms of a hazard function (the risk of instantaneous death at a given time assuming survival to that time). The hazard function is in turn a function of the presumed prognostic factors and their respective regression coefficients. The validity of the assumption of proportionality of hazards was examined by visual inspection of individual Weibull plots [log(-log-(Surv(t))) versus log (t)] depicting the survival of patients grouped according to whether their respective values were below or above the median value for the variable being tested. If the assumption of proportionality of hazards holds for a covariate, the plots for patient groups with values above and below the median should be parallel. In general, no gross deviations from the parallel were observed.

Results. Patient population and survival outcomes. Fifty patients (20.4%) died during the 9-month trial. The baseline characteristics of the survivors were compared with those who died(Table). The mean baseline values of several variables were different for the survivors than for the patients who died, including bulbar onset of ALS, total ALSFRS score, body weight, and combined muscle strength megascore at baseline. However, baseline ALSFRS, isometric muscle strength, and body weight were not significant predictors of survival in the Cox model. Because isometric muscle strength has been considered to be a fundamental measure of disease progression^10,11 and was the primary end point for the trial, we included it in the model presented here. However, removing isometric strength from the model does not affect the risk ratios for the remaining variables. (The effect of stepwise removal of certain baseline measures, including isometric muscle strength, is available through NAPS file number 05436.)

Mean FVC% declined linearly during the treatment period.⁹ To examine how FVC% changed in the months before death, for each patient who died either during the active treatment or extended observation period, data were plotted backward in time from the date of death. Mean FVC% declined roughly linearly with no apparent inflection point to indicate an impending terminal event (figure 1A). This back-extrapolated graph shows that whereas the mean FVC% for the group declined approximately linearly up until 4 to 5 months before death, it remained relatively stable in the range of 20 to 35% within the final 3 months of illness. The use of mechanical ventilation cannot account for this flattening in the curve, because less than 10% of patients opted for external ventilatory support; only 12 patients used bilevel positive airway pressure(BIPAP) and 5 were placed on invasive mechanical ventilation during the blinded period of observation in this study.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1. Plots of (A) mean (± SE) change in FVC%, (B) serum CO₂, and (C) serum chloride in the months preceding death. These plots include all patients who died during the combined 9-month blinded study plus open-label extension follow-up period.

We therefore sought to determine if other physiologic measures might reflect ongoing respiratory decompensation during the preterminal phase of the illness when FVC% remained relatively stable but poor. We hypothesized that marginal pulmonary function might be reflected in the development of a respiratory acidosis with concurrent metabolic compensation. Taking advantage of the fact that monthly safety evaluations performed during this pharmaceutical clinical trial included regular assessment of serum chemistries, we examined the behavior of serum CO₂ and serum chloride values in the months before death.

Average values for serum CO₂ started to rise continuously at about 9 to 10 months before death (figure 1B). In contrast, mean serum chloride declined slowly at first but then dropped sharply starting at about 4 to 5 months before death. The mean value dropped below the clinical laboratory's normal lower limit (data on file, SmithKline Beecham Clinical Laboratories, Van Nuys, CA) only in the final month before death (figure 1C).

Baseline variables as predictors of death. For placebo patients who entered the treatment trial, significant baseline predictors of 9-month survival included age, FVC% and serum chloride at study entry, bulbar onset of ALS, the time from symptom onset to diagnosis, and weight loss observed during the 2- to 6-month screening period before commencement of the active treatment portion of the study. Neither the baseline combined limb muscle strength (assessed by maximal voluntary isometric contraction)⁽¹⁰⁾ nor the rate of change in muscle strength during the screening study was a significant predictor of survival(figure 2).

• Download figure
• Open in new tab
• Download powerpoint

Figure 2. The 95% confidence intervals for the risk ratio, or average change in survival of ALS patients for unit change, of each prognostic factor. A relative risk of "1" indicates that a given factor is not prognostically related to outcome. A positive risk ratio indicates that increasing values of the factor are associated with an increased mortality risk; negative values signify that decreasing values of the parameter are associated with increased risk. If the confidence interval includes 1 (i.e., the open boxes representing the 95% confidence limits for a factor depicted on the graph cross the center line which denotes a relative risk of 1, as for example the baseline muscle strength measure), then that factor is not statistically significant.

The predictive equation based on this model (assuming 80% mean survival) is Equation 1. where P is the survival probability at 9 months, A is age (years) at the start of observation interval, D is months from first reported ALS symptom to formal diagnosis, F is baseline FVC%, C is baseline serum chloride (mEq/L), W is the prestudy weight change (percent per month), S is the baseline isometric muscle strength megascore(Z-units¹⁰), and B is bulbar onset (0, no; 1, yes). Parameter estimates and their respective 95% confidence limits for this survival model are presented in table of the archived data supplement to this article (see NAPS document no. 05436). These parameter estimates are point estimates obtained from observed data. The variability of this predictive equation is a function of the standard errors of all parameters; it has not been formally quantified using simulated data. The observed data appear consistent with the assumption of proportionality of hazards for all of the covariates in the model.

Other exploratory analyses. Additional Cox proportional hazards models were examined to determine which factors were the strongest predictors of survival. Specifically, we asked if the baseline values of FVC% and chloride were better predictors than the last observed values or whether the rates of change were stronger predictors. In some models, the weaker factors(age, time from symptom onset to diagnosis, screening study weight loss) were not significant at the 0.05 level. Baseline FVC%, baseline chloride, and their respective slopes were consistently stronger predictors (see NAPS document no. 05436).

A logistic regression model was used to express the relationship between FVC% at study entry and 9-month survival probability.Figure 3A shows that for this group of patients, 9-month predicted survival probability declined rapidly for patients with FVC% below a baseline value of 60%. Similarly,figure 3B shows the relationship between survival probability and baseline serum chloride.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3. Nine-month survival probabilities based on baseline FVC% (A) or baseline serum chloride (B) values. The plotted probabilities are the predicted values resulting from logistic regression fits of survival outcome (alive or dead at month 9) as a function of each respective prognostic factor alone.

Discussion. We evaluated variables that might be used to predict survival outcome in ALS, with a particular view toward their use in clinical trials. Pulmonary function as assessed by FVC% and serum chloride measurements have great predictive value. This abrupt change in the rate of decline in serum chloride, which reflects a metabolic compensation for the chronic respiratory acidosis that occurs as pulmonary function deteriorates in the later stages of ALS, may reflect the impending failure of mechanisms of respiratory compensation.

In other diseases, survival prediction methods have been used to help clinicians identify prognostic markers to aid in patient management. For example, in chronic obstructive pulmonary disease, forced expiratory volume in 1 second (FEV₁) has been documented as a strong long-term predictor of mortality.¹² However, when a patient's disease has progressed to long-term oxygen therapy, FEV₁ is not as strong a predictor as pretherapy pulmonary artery mean pressure (PAP), a marker for pulmonary hypertension.¹³ In the study of Oswald-Mammosser et al.,¹³ patients were included if their FEV₁ values were <60% of normal, but the range of observed baseline FEV₁ and FEV₁/FVC values indicated moderate to severe respiratory impairment.¹⁴ Even though long-term oxygen therapy tends to stabilize or reverse pulmonary hypertension, PAP measured at onset of therapy serves as an indicator of overall disease severity. By analogy in ALS, when a patient's FVC% reaches 35% or lower, it is apparent that the patient is in the terminal stage of the disease, but it is unclear if and how quickly it may decline further, making FVC% hard to use as a prognostic marker. In addition, FVC% may be difficult to test accurately in the later stages of the disease, because to do so requires specialized equipment, intact facial muscle strength and coordination, in addition to repeated maximal voluntary efforts by the patient.

Our observations have distinct implications for the design and conduct of clinical trials and for patient management in ALS.

From the standpoint of designing clinical trials in which survival is the primary outcome measure, we note that a patient's probability of surviving over a given interval is a function of the FVC% value at the start of that interval. We now can describe the statistical relationship between FVC% and survival probability in terms of reduction of mortality risk. The ability of a treatment intervention to disrupt the relationship we have described between baseline FVC%, serum chloride, and 9-month survival should be reflected in a change in the hazard function. Thus, FVC% and serum chloride, along with the patient's age, are strong prognostic indicators that can be used to guide patient selection and that should also be included as covariates in analyses of outcome. We must caution, however, that our observations were obtained in the setting of a pharmaceutical clinical trial. Thus, despite the broad inclusion criteria, factors such as patient selection, increased vigilance by physicians and caregivers, and patients' expectations may have influenced the mathematical relationships we deduced in our model. The precise parameter estimates we derived may not, therefore, apply to the ALS population as a whole. Nonetheless, the conceptual relationships we discerned are robust and are congruent with clinical experience, lending general utility to the model.

For clinicians following patients' progress over time, the change in serum chloride values and the point when these values fall below the lower limit of normal are sensitive indicators of impending respiratory decompensation and thus appear to constitute grave prognostic signposts in the clinical course of ALS. A serum chloride value is easily obtainable, even for those patients in home care and/or hospice settings where objective assessment of pulmonary function cannot be accurately obtained.

Acknowledgments

We thank Dr. Janet Wittes for her thoughtful statistical advice and comments. We also thank the members of the A.C.T.S. group for their careful collection of the clinical data, which comprise this work.

Appendix.